Wrapper-Based Feature Selection Using Sperm Swarm Optimization: A Comparative Study

Feature selection is a vital technique that enhances the quality of input datasets by reducing redundancy, noise, and inaccuracies without compromising classifier accuracy. The integration of metaheuristic search algorithms (MSAs) into feature selection enables the discovery of relevant features, thereby simplifying dataset representation. However, despite the proliferation of MSAs based on diverse sources of inspiration in recent years, their application to solve real-world optimization problems, such as feature selection, remains largely unexplored. This paper introduces a novel MSA called sperm swarm optimization (SSO), inspired by the mobility behavior of sperms during the fertilization process, to develop a robust wrapper-based feature selection method. To evaluate the effectiveness of SSO in solving feature selection problems, extensive simulations are conducted using ten datasets from the UCI Machine Learning Repository. The results are compared with five other prominent MSAs. Extensive simulations demonstrate that SSO outperforms the peer algorithms in terms of mean accuracy and feature selection efficiency across the majority of the employed datasets. The experimental results provide compelling evidence for the efficacy and potential of SSO in addressing feature selection challenges.